the effects of structural reforms on productivity and profitability …adkugler/jde_ehkk.pdf ·...
Post on 09-Jan-2020
8 Views
Preview:
TRANSCRIPT
Journal of Development Economics 75 (2004) 333–371
www.elsevier.com/locate/econbase
The effects of structural reforms on productivity
and profitability enhancing reallocation:
evidence from Colombia
Marcela Eslavaa, John Haltiwangerb,
Adriana Kuglerc, Maurice Kuglerd,*
aUnibersidad de Los Andes, Bogota, ColombiabUniversity of Maryland and NBER, USA
cUniversity of Houston, UK; Universitat Pompeu Fabra, and CREA, Spain, CEPR and IZAdEconomics Division, School of Social Sciences, University of Southampton,
Southampton SO17 IBJ, United Kingdom
Abstract
Estimates for the U.S. suggest that at least in some sectors productivity enhancing reallocation is
the dominant factor in accounting for productivity growth. An open question, particularly relevant
for developing countries, is whether reallocation is always productivity enhancing. It may be that
imperfect competition or other barriers to competitive environments imply that the reallocation
process is not fully efficient in these countries. Using a unique plant-level longitudinal dataset for
Colombia for the period 1982–1998, we explore these issues by examining the interaction between
market allocation, and productivity and profitability. Moreover, given the important trade, labor and
financial market reforms in Colombia during the early 1990s, we explore whether and how the
contribution of reallocation changed over the period of study. Our data permit measurement of plant-
level quantities and prices. Taking advantage of the rich structure of our price data, we propose a
sequential methodology to estimate productivity and demand shocks at the plant level. First, we
estimate total factor productivity (TFP) with plant-level physical output data, where we use
downstream demand to instrument inputs. We then turn to estimating demand shocks and mark-ups
with plant-level price data, using TFP to instrument for output in the inverse-demand equation. We
examine the evolution of the distributions of TFP and demand shocks in response to the market
0304-3878/$ -
doi:10.1016/j.
* Corresp
E-mail add
adkugler@uh.
see front matter D 2004 Elsevier B.V. All rights reserved.
jdeveco.2004.06.002
onding author.
resses: eslava@econ.umd.edu (M. Eslava)8 haltiwan@econ.umd.edu (J. Haltiwanger)8
edu (A. Kugler)8 maurice.kugler@soton.ac.uk (M. Kugler).
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371334
reforms in the 1990s. We find that market reforms are associated with rising overall productivity that
is largely driven by reallocation away from low- and towards high-productivity businesses. In
addition, we find that the allocation of activity across businesses is less driven by demand factors
after reforms. We find that the increase in aggregate productivity post-reform is entirely accounted
for by the improved allocation of activity.
D 2004 Elsevier B.V. All rights reserved.
Keywords: Structural reforms; Reallocation; Productivity growth; Demand shock vs. technical efficiciency
decomposition
1. Introduction
Market economies are continually restructuring in response to changing conditions. The
burgeoning evidence from longitudinal micro business databases has shown that
productivity growth at the aggregate level is closely connected to the efficiency of the
economy at the micro level to allocate outputs and inputs across businesses. That is, a large
fraction of measured productivity growth is accounted for by more productive expanding
businesses displacing less productive contracting businesses.1
These issues loom especially large in developing economies—although, they have hardly
been settled in advanced economies. In developing economies, there are potentially a variety
of barriers to promoting efficient ongoing reallocation. These barriers might stem from
distortions in market structure as well as the market institutions and policies in place. Aware
of the impediments that distortionary policies may place on reallocation and, more generally,
on productivity growth, during the past decade Latin American countries have undertaken a
whole series of reforms (including, labor, financial, and trade reforms) to promote flexibility.
In this paper, we characterize the evolution of the distributions of productivity and other
determinants of profitability such as demand factors in Colombia, one of the Latin
American economies that underwent extensive structural reforms in the early nineties. The
role of productivity vs. other components of profitability is critical in evaluating the
market-oriented reforms in Latin America since the reforms were intended to enhance the
role of productivity and perhaps reduce the role of demand factors (especially to the extent
to which markets were artificially imperfectly competitive) in the allocation of activity. We
examine the relationship between the allocation of activity, on the one hand, and
productivity and demand factors on the other.2 We then ask how this relationship changed
after market reforms were introduced in Colombia in the early 1990s.
Colombia is a superb country to study these issues for two reasons. First, Colombia
underwent a substantial and relatively fast market reform process, mainly in 1990 and
1991. The 1990 labor market reform, which introduced individual severance payments
1 For the labor market, the efficient churning of businesses implies a need to reallocate jobs at a high pace
without long and costly spells of unemployment. The latter is important for not only productive efficiency of the
economy but also obviously important for welfare.2 In our analysis, we focus on some of the main determinants of profitability, including productivity, demand
and output prices. There are other determinants of profitability, such as costs, which are not the focus of our
present analysis.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 335
savings accounts, reduced dismissal costs by between 60% and 80% (Kugler, 1999, 2004).
The 1991 trade liberalization reduced the average tariff from over 62% to around 15%
(Lora, 1997). The financial reform first introduced in 1990 and extended in 1991,
liberalized deposit rates, eliminated credit subsidies and modernized capital market and
banking legislation (Lora, 1997). In addition, restrictions on inflows of foreign direct
investment (FDI) were removed in 1991 as multinational firms were given national
treatment (Kugler, 2000). The 1993 social security reform allowed voluntary transfers
from a pay-as-you-go system to a fully funded system with individual accounts, though it
increased employer and employee pension contributions up to 13.5% of earnings (Kugler
and Kugler, 2003).
Second, Colombia has unique longitudinal microeconomic data on businesses. A
unique feature of the Colombian data is that both plant-level quantities and prices can be
measured. The ability to measure plant-level prices of both outputs and inputs is
potentially very important in this context for both measurement and conceptual reasons.
Much of the existing literature measures establishment output as revenue divided by a
common industry-level deflator. Therefore, within-industry price differences are embodied
in output and productivity measures. Moreover, if prices reflect idiosyncratic demand
shifts or market power variation rather than quality or other differences in product
attributes, then high bmeasuredQ productivity businesses may not be particularly efficient.
In this sense, the relationship between productivity and reallocation in the literature may
be misleading.3
In the context of a developing economy undergoing structural reforms these
measurement and conceptual issues are especially important. A key objective of
structural reforms is to make markets more competitive. Without the ability to measure
plant-level prices it is very difficult to quantify and, in turn, analyze the respective
contributions of demand and efficiency factors in the allocation of activity and how
they may have changed in response to market reforms. In this paper, we attempt to
measure the separate influences of idiosyncratic productivity and demand on the
allocation of activity in Colombia.
In exploiting these unique micro data, our paper makes a number of related
methodological innovations relative to much of the literature. First, in measuring
efficiency we estimate production functions using an extension of the instrumental
variable approach pioneered by Shea (1993) and Syverson (2003). In the latter, Syverson
uses local downstream demand instruments to estimate returns to scale of production
functions. Building on this approach, we use downstream demand instruments as well as
plant-level input prices for materials and energy as instruments in estimating production
functions. From this structural estimation of production functions, we generate measures
3 Since Marschak and Andrews (1944), there has been awareness about the possible difficulties involved in
using revenue-based deflated output in establishment level data. Klette and Griliches (1996) consider how intra-
industry price fluctuations can affect production function and productivity estimates. Melitz (2003) explores this
problem further and extends the analysis to consideration of multi-product producers. Katayama et al. (2003)
argue that both revenue-based output and expenditure-based input measures can lead to productivity
mismeasurement and incorrect interpretations about how heterogeneous producers respond to shocks and the
associated welfare implications. Foster et al. (2003) use plant-level data on quantities and prices for the U.S. to
study market selection dynamics.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371336
of plant-level efficiency (i.e., plant-level total factor productivity). Second, since we can
measure plant-level output prices, we estimate demand elasticities using the total factor
productivity estimates as instruments in the demand equations. That is, we first use
downstream demand instruments to estimate production functions and then use supply
shocks (i.e., total factor productivity) to estimate the demand functions.
Having estimated production and demand functions, we study the evolution of the
distributions of total factor productivity, demand shocks, and prices through the market
reforms in Colombia. We find that aggregate TFP is higher after the reforms. Moreover,
the dispersion of TFP also rises after the reforms. Our main focus though is on the
allocation of activity and its relationship to TFP and demand shocks in Colombia
throughout the entire period as well as during periods prior to and subsequent to the
reforms. Overall, we find that the allocation of activity reflects both efficiency and demand
factors. That is, market shares are higher for businesses that are more efficient and face
higher demand. Following the reforms, we find that efficiency factors become more
important in the allocation of activity across businesses while demand factors become less
important. Interestingly, we find that the rising aggregate productivity in Colombia over
this period is accounted for almost entirely by the improved allocation of activity across
businesses.
At the extensive margin, we also find that exiting businesses are less efficient and face
lower demand for their products relative to incumbents. On the other hand, entering
businesses are more efficient than incumbents and exiting businesses, but also face lower
demand relative to incumbents. This finding points to the importance of vintage
technology rather than learning to account for productivity dynamics. Note that
revenue-based TFP measures spuriously include a demand component and thus under-
estimate technical efficiency of entrants, who have to build a consumer base, relative to
incumbents, who have a established clientele. Hence, the analysis in the present paper
provides a better basis to discern between vintage and learning explanations for dynamics
and heterogeneity in productivity.
How did these patterns change in response to the market reforms? After the market
reforms, the positive gap between the productive efficiency of the average incumbent and
the average exiting business rises but the positive gap between the productive efficiency of
the average entering business and the average incumbent declines. On net, the contribution
of entry and exit to average productivity rises slightly. Moreover, when we investigate a
more dynamic specification we find that, after the reforms, entrants exhibit slightly higher
productivity than incumbents. There is also evidence of learning effects in the post-reform
period. Putting these pieces together suggests market reforms yielded a more heteroge-
neous group of entrants but that conditional on survival the post-reform entrants contribute
more to productivity. Increased learning by entering businesses after the reforms may be
due to increased access to new imported capital vintages and increased access to know-
how from foreign businesses. These patterns are also consistent with greater selection by
the market of the least productive and profitable businesses after foreign competition was
increased, access to subsidized credit was reduced, and dismissal costs were lowered
following the reforms.
Before proceeding, it is useful to emphasize that the welfare implications (and even the
productivity implications) are complex in this context with imperfect competition. Our
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 337
paper is intended to be a contribution on characterizing the positive implications of market
reform rather than the normative implications. There are several interesting complicating
factors in considering welfare implications. For one, the market power of firms may derive
from economic fundamentals or from institutional factors. For example, it may be that
market power derives from product differentiation by each producer and that markets are
best characterized by monopolistic competition. The exploitation of market power in this
context by firms still generates a welfare loss with prices being above marginal cost, but
even in the absence of markups substantial price heterogeneity across firms would still be
a feature of a first-best equilibrium. Alternatively (or in addition), market power may
derive from institutional barriers (e.g., trade restrictions). Simple intuition suggests that
elimination of such barriers will improve welfare, but even here one should be cautious
given potential second-best considerations. Another related complicated factor is that
market power may play an important role in product and process innovations that lead to
technological progress. In some creative-destruction models of innovation and growth
(e.g., Aghion and Howitt, 1992), market power plays an important role for innovators in
an environment in which there are fixed costs of innovation. We do not have much to say
directly about this latter class of models since we do not directly investigate the role of
product and process innovation. Part of the reason for this is that our empirical focus is on
operating plants in manufacturing (not R&D facilities). Our analysis has more to say about
heterogeneity resulting from differences in adoption rather than innovation. However, a
long-standing issue in the empirical innovation literature is how much innovation occurs in
R&D facilities and how much occurs on the factory floor.
The rest of the paper proceeds as follows. Section 2 describes the Colombian context
and the market reforms undertaken during the period of study. Section 3 describes the
unique data we use in this analysis. Section 4 describes our methodology for estimating
plant-level total factor productivity and demand shocks, and presents the results of these
estimations. Section 5 describes basic patterns of productivity, demand, and prices in terms
of dispersion and persistence. Section 6 examines the relations between efficiency and
demand, on the one hand, and the allocation of activity, on the other hand, both before and
after the reforms. Section 7 provides concluding remarks.
2. Market reforms in Colombia
In 1990, the administration of President Cesar Gaviria conceived a comprehensive
reform package, which included not only measures to modernize the state and liberalize
markets but also a constitutional reform. In contrast to the experience in many reformist
countries, there was no underlying economic crisis preceding the reform in Colombia.
Rather, ample social and political unrest were the catalysts for institutional change (see,
e.g., Edwards, 2001). Structural reforms to remove distortions from product and factor
markets were introduced in parallel with constitutional changes to political institutions as
part of a wider effort to control internal strife associated with violent activities of both
illicit drug cartels and guerrilla groups. The absence of an economic crisis meant that there
was no clear consensus for the need of market-oriented reforms, but at the same time good
economic conditions were initially an asset as distributive conflict was addressed by
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371338
elaborate resource transfer schemes (see, e.g., Kugler and Rosenthal, 2004). However, the
inability to reconcile market oriented reforms with the continued and pervasive need for
redistribution to compensate losers meant that by 1993 the spurt of reform of the early
nineties came to a halt. Although there was no reversal of reforms, Colombia did not make
additional progress in terms of further removing distortions while many other countries in
Latin America, and around the world, actually deepened reform efforts (see, e.g., Burki
and Perry, 1997). The fact that the structural reform process in Colombia was a one-off
phenomenon aids our identification strategy in trying to assess its impact on the relation
between reallocation and productivity.
Before Gaviria’s administration, the government of President Virgilio Barco made some
partial progress in trade liberalization but did not gain any significant ground in removing
other distortions. The gradual decrease in tariffs initiated by Barco was accelerated by
Gaviria after June 1991. By the end of 1991, nominal protection reached 14.4% and
effective protection 26.6%, down from 62.5% a year earlier, while 99.9% of items were
moved to the free import regime. These measures clearly generated unrest among the
owners of capital, who faced lower profit margins after trade liberalization due to
increased foreign competition. The impact from market penetration by foreign exporters,
and by foreign investors as pointed out below, on the market participation of domestic
businesses in product markets was mitigated by more favorable conditions in factor
markets, which lowered both capital and labor costs. In labor markets, Law 50 of
December 1990 promoted the flexibilization of contracts and reduced labor costs by
between 60% and 80% (see, e.g., Kugler, 1999, 2004). In 1990, the government also tried
to introduce changes in the social security system as part of the labor reform package, but
Congress forced an independent process to reform pension provision. The opposition to
the original reform plan by the Executive stemmed partly from the proposal to lower labor
costs by reducing employer’s contributions. Later during the Gaviria administration, the
executive compromised with Congress by passing Law 100 in 1993. Although Law 100
allowed for voluntary individual conversion from a pay-as-you-go system to a fully funded
system with accounts, this law also introduced a mandatory hike in employer and
employee contributions up to 13.5% of salaries, of which 75% was paid by employers
(see, e.g., Kugler and Kugler, 2003).
A number of measures also reduced frictions in financial markets. In 1990, Law 45 was
introduced with the goal of reducing state control and ownership concentration in the
banking sector. Interest rate ceilings were eliminated and required reserves were reduced.
At the same time, supervision was reinforced in line with the Basle Accords for
capitalization requirements, though there were no requirements for banks to invest in
government securities. Law 9 of 1991 established the abolition of exchange controls
eliminating the monopoly of the central bank on foreign exchange transactions and
lowering substantially the extent of capital controls. Finally, Resolution 49 of 1991
eliminated restrictions to foreign direct investment. This resolution established national
treatment of foreign enterprises and eliminated limits on the transfer of profits abroad as
well as bureaucratic procedures requiring the approval of individual projects (see, e.g.,
Kugler, 2000). This measure not only facilitated capital inflows across all sectors, but also
induced entry of foreign banks increasing competition in the financial sector, thereby
yielding lower intermediation costs.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 339
After the end of Gaviria’s term, in 1994 President Ernesto Samper gained power on
a platform partially based on opposition to market institutions.4 While the new
government was unsuccessful in terms of dismantling existing reforms, the coalition
supporting the new government managed to bring the momentum for additional reforms to
a halt.
Overall, though, the reform process was relatively smooth and successful in
transforming a highly distorted economy into one burdened with many fewer frictions.
Trade liberalization and the removal of FDI restrictions were aimed at increasing
competition and shifting production away from low- and towards high-productivity
businesses. In addition, the liberalization of factor markets had the purpose of
facilitating reallocation in the economy after the structural reform process. If the goal of
inducing productivity-enhancing reallocation was achieved, its success should be
reflected in different patterns of industrial dynamics between the 1980s and the 1990s,
with market allocation being driven to a large extent by productivity rather than
profitability in the latter period. In the following sections, we investigate whether this is
the case.
3. Data description
Our data come from the Colombian Annual Manufacturers Survey (AMS) for the years
1982 to 1998. The AMS is an unbalanced panel of Colombian plants with more than 10
employees, or sales above a certain limit (around US$35,000 in 1998).5 The AMS includes
information for each plant on: value of output and prices charged for each product
manufactured; overall cost and prices paid for each material used in the production
process; energy consumption in physical units and energy prices; production and non-
production number of workers and payroll; and book values of equipment and structures.
The dataset also provides information on plant location as well as industry classification
codes (five digits CIIU).
Our aim in this paper is to estimate and describe the behavior of productivity and
demand shocks in Colombia during the pre- and post-reform periods and to document how
reallocation contributed to productivity and demand increases during the period of study
and, especially, whether the relation changed after the reforms. To this end, we estimate
total factor productivity (TFP) values for each plant using a capital–labor–materials–
energy (KLEM) production function and demand shock values for each plant using a
standard inverse-demand function. To estimate production and inverse-demand equations,
we need to construct physical quantities and prices of output and inputs, capital stock
series, and total labor hours.
4 Note that the Colombian electoral system rules out re-election.5 In Appendix A, we describe the methodology used to generate longitudinal linkages that allow following
plants over time. As explained in Appendix A, over the period of study the AMS underwent changes in the
sampling and labeling of plants, which require very involved work to generate these linkages and reduce spurious
measurement of entry and exit of plants.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371340
3.1. Plant-level price indices
With the rich information on prices collected in the AMS, we can construct plant-level
price indices for output, materials, and energy. This represents an enormous advantage
with respect to other sources of data, as the use of more aggregate price deflators is a
common source of measurement error, due to plant-specific demand shocks. Prices of
output and materials are constructed using Tornqvist indices. For a composite of products
or materials i of each plant j at time t that are constructed these Tornqvist indices are the
weighted average of the growth in prices for all individual products or materials h
generated by the plant. This weighted average is given by:
DPijt ¼XHh¼1
shjtDln Phjt
�;
�
where i=Y, M, i.e., output or materials,
Dln Phjt
�¼ lnPhjt � lnPhjt�1;
�
and
shjt ¼shjt þ shjt�1
2
and Phjt and Phjt�1 are the prices charged for product h, or paid for material h, by plant j at
time t and t�1, and sht and sht�1 are the shares of product h in plant j’s total production, or
the shares of material h in the total value of plant j’s materials purchases, for years t and
t�1.6 The indices for the levels of output (or material) prices for each plant j are then
constructed using the weighted average of the growth of prices and fixing 1982 as the base
year:
lnPjt ¼ lnPijt�1 þ DPijt
for t N1982, where Pj1982=100, and where the price levels are then simply obtained by
applying an exponential function to the natural log of prices, Pijt=explnPijt.7
3.2. Capital stock series
Given prices for materials and output, the quantities of materials and output are
constructed by dividing the cost of materials and value of output by the corresponding
prices. Quantities of energy consumption are directly reported by the plant. In addition, we
need capital stocks to estimate a KLEM production function.
6 The distribution of the weighted average of the growth of prices has large outliers, especially at the left side
of the distribution. In particular, the distribution shows negative growth rates of 100% and more. In a country like
Colombia, with inflation around 20%, negative growth rates of these magnitudes seem implausible. For this
reason, we trim the 1% tails at both ends of the distribution as well as any observation with a negative growth rate
of prices of more than 50%.7 Given the recursive method used to construct the price indices and the fact that we do not have plant-level
information for product and material prices for the years before plants enter the sample, we impute product and
material prices for each plant with missing values by using the average prices in their sector, location, and year.
When the information is not available by location, we input the national average in the sector for that year.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 341
The plant capital stock is constructed recursively on the basis of the expression:
Kjt ¼ 1� dð ÞKjt�1 þIjt
Dt
for all t such that Kjt�1N0, where Ijt is gross investment, d is the depreciation rate and Dt is
a deflator for gross capital formation. Our measure of Dt is the implicit deflator for capital
formation from the input–output matrices for years 1991–1994, and from the output
utilization matrices for later years. We use the bobservedQ depreciation rates calculated in
Pombo (1999) at the three-digit sectoral level, which range between 8.7% and 17.7% for
machinery and between 2.4% and 9.8% for buildings.
Gross investment is generated from the information on fixed assets reported by each
plant, using the expression:
Ijt ¼ KNFjt � KNI
jt þ djt � pAjt
where KjtNF is the value of fixed assets at the end of year t reported by plant j, Kjt
NI is the
value of fixed assets reported by plant j at the start of year t, djt is the depreciation reported
by plant j at the end of year t, and pitA is the reported inflation adjustment to fixed asset value
by plant j at the end of year t (only relevant since 1995, the first year in which plants were
required by law to consider this component in their calculations of end-of-year fixed assets).
The capital stock series is initialized at the year the plant enters the sample, t0. To obtain
the initial value, we use the equation:
Kit0 ¼KNIit0
0:5Dt0 þ 0:5Dt0�1
;
where Kjt0
NI is the first reported nominal capital stock at the beginning of the year. In
constructing the capital stock series we include equipment, machinery, buildings and
structures, while excluding vehicles and land.
3.3. Total labor hours
Finally, we construct a labor measure as total hours of employment. Since the AMS
does not have data on production and non-production worker hours, we construct a
measure of total employment hours at time t for sector G( j), to which plant j belongs, as,
hG jð Þt ¼earningsG jð Þt
wG jð Þt;
where wG( j)t is a measure of sectoral wages at the three-digit level from the Monthly
Manufacturing Survey, and earningsG( j)t is a measure of earnings per worker constructed
from our data as
earningsG jð Þt ¼P
jaG payrolljtPjaG ljt
Table 1
Descriptive statistics for continuing, entering, and exiting plants, before and after market-oriented reforms
Variable Pre-reforms Post-reforms
All Continuers Entrants Exits All Continuers Entrants Exits
Output 10.49
(1.67)
10.72
(1.66)
9.47
(1.29)
9.39
(1.36)
10.90
(1.88)
11.12
(1.83)
9.87
(1.64)
9.72
(1.60)
Capital 8.21
(2.05)
8.42
(2.04)
6.93
(1.66)
7.30
(1.81)
8.75
(2.18)
8.91
(2.15)
7.51
(1.91)
7.76
(1.96)
Labor 10.97
(1.1)
11.08
(1.10)
10.21
(0.78)
10.32
(0.83)
10.95
(1.25)
11.13
(1.23)
10.15
(1.07)
10.30
(1.11)
Energy 11.30
(1.88)
11.51
(1.89)
10.16
(1.42)
10.34
(1.55)
11.55
(1.99)
11.76
(1.97)
10.45
(1.71)
10.64
(1.75)
Materials 9.61
(1.85)
9.83
(1.83)
8.58
(1.54)
8.48
(1.57)
10.25
(1.88)
10.45
(1.84)
9.32
(1.66)
8.99
(1.71)
Output prices �0.08
(0.44)
�0.09
(0.46)
�0.07
(0.51)
�0.06
(0.47)
�0.15
(0.74)
�0.15
(0.71)
�0.10
(0.8)
�0.09
(0.81)
Energy prices 0.25
(0.50)
0.28
(0.52)
0.32
(0.55)
0.22
(0.49)
0.55
(0.43)
0.55
(0.41)
0.61
(0.49)
0.55
(0.41)
Material prices 0.02
(0.35)
0.02
(0.37)
0.05
(0.43)
0.02
(0.38)
�0.10
(0.57)
�0.09
(0.53)
�0.13
(0.71)
�0.01
(0.73)
Entry rate 0.0981 0.0843
Exit rate 0.0873 0.1069
N 55,298 41,265 4846 4826 44,816 32,395 3778 4202
This table reports means and standard deviations of the log of quantities and of prices deviated from yearly
producer price indices. The entry and exit rates are the number of entrants divided by total plants and number of
exiting firms divided by total number of plants. The pre-reform period includes the years 1982–1990 and the post-
reform period includes the years 1991–1998.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371342
Then, the total employment hours measure is constructed as,8
Ljt ¼ ljthG hð Þt;
where ljt is the total number of employees in firm j at time t.
3.4. Descriptive statistics
Table 1 presents descriptive statistics of the quantity and price variables just described,
for the pre- and post-reform periods, by continuing, entering or exiting status. The table
reports entry and exit rates of 9.8% and 8.7% during the pre-reform period, but a lower
entry rate of 8.4% and a higher exit rate of 10.7% during the post-reform period.9 The
quantity variables are expressed in logs, while the prices are relative to a yearly producer
price index to discount inflation. Output increased between the pre- and post-reform
periods. Similarly, except for labor, input use increased between the pre- and post-reform
8 While we observe total employment at the plant-level, we do not observe total hours. Our method is to
impute total hours per worker for plant j using the average for plants in the same industry as plant j. A large share
of the variation in total employment hours stems from total employment variation.9 These entry and exit rates are lower than those reported for the U.S. and other OECD countries (Davis et al.,
1996). Given that the Colombian economy is subject to greater rigidities, one may indeed expect lower entry and
exit rates in the Colombian context (see, e.g., Tybout, 2000 for a discussion of this issue).
Table 2
Correlations of variables
Output Capital Labor Energy Materials Output
prices
Energy
price
Materials
prices
Output 1.0
[100,114]
0.7550
[96,232]
0.7393
[99,102]
0.7649
[99,476]
0.8982
[90,938]
�0.2918
[100,114]
�0.0591
[100,095]
�0.0705
[91,540]
Capital 1.0
[96,232]
0.6904
[95,303]
0.7592
[95,684]
0.7267
[87,519]
�0.0092
[96,232]
�0.0082
[96,213]
�0.0241
[88,066]
Labor 1.0
[99,102]
0.6963
[98,484]
0.6761
[90,073]
0.0084
[99,102]
�0.0389
[99,083]
0.0067
[90,671]
Energy 1.0
[99,476]
0.7492
[90,394]
0.0087
[99,476]
�0.1198
[99,457]
�0.0188
[90,944]
Materials 1.0
[90,938]
�0.0063
[90,938]
0.0207
[90,929]
�0.2384
[90,938]
Output prices 1.0
[100,114]
�0.0047
[100,095]
0.2469
[91,540]
Energy prices 1.0
[100,095]
�0.0118
[91,531]
Materials prices 1.0
[91,540]
The table includes bivariate correlations of logs of quantities and price deviations from yearly producer price
indices. The numbers in square brackets are the number of observations.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 343
periods. In particular, the table shows that capital, materials and energy use increased
between the pre- and post-reform periods especially for incumbents and entrants. On the
other hand, labor use decreased between the two periods for entering and exiting plants.
Relative prices of output and materials prices declined between the pre- and post-reform
periods for all plants.10
Table 2 reports simple correlations of the various variables reported in Table 1, which
display by and large the expected patterns. Output is positively correlated with all inputs,
negatively correlated with its own price, and negatively but weakly correlated with energy
and materials prices. Inputs are positively correlated with each other. All inputs are
negatively correlated with energy prices, except for materials. All inputs are negatively
correlated with materials prices, except for labor. Energy and materials uses are negatively
correlated with their own prices. Also, as expected, output prices are positively correlated
with materials prices, and negatively but weakly correlated with energy prices. Finally,
energy and materials prices are weakly and negatively correlated.
In the next section, we use these variables to estimate the production function and
inverse-demand equation.
10 Caution needs to be used in interpreting the aggregate (mean) relative prices in this context since the
relative price at the micro level is the log difference between the plant-level price and the log of the aggregate PPI.
On an appropriately output weighted basis, the mean of this relative price measure should be close to zero in all
periods (or one in levels) since the PPI is dominated by manufacturing industries. The larger difference with
respect to PPI in the post-reform period reflects that the growth of manufacturing prices fell more rapidly than that
of other prices in the economy, possibly due to the fact that external competition introduced by the reforms
affected the manufacturing sector more than others.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371344
4. Estimation of TFP and demand shocks
We estimate total factor productivity with plant-level physical output data, using
downstream demand to instrument inputs. In turn, we estimate demand shocks and mark-
ups with plant-level price data, using TFP to instrument for output in the inverse-demand
equation.
4.1. TFP estimation
We estimate total factor productivity for each establishment as the residual from a
capital–labor–materials–energy (KLEM) production function:
Yjt ¼ KajtL
bjtE
cjtM
/jt Vjt;
where Yjt is output, Kjt is capital, Ljt is total employment hours, Ejt is energy consumption,
Mjt are materials, and Vjt is a productivity shock. We estimate this production function in
logs,
logYjt ¼ alogKjt þ blogLjt þ clogEjt þ /logMjt þ logVjt;
so that our total factor productivity measure is estimated as:
TFPjt ¼ logYjt � aalogKjt � bblogLjt � cclogEjt � //logMjt: ð1Þ
where a, b,c, and u are the estimated factor elasticities for capital, labor, energy, and
materials. In addition, we estimate a value-added production function with only capital and
employment to provide a benchmark of comparison with other studies. Whether we
estimate value-added or KLEM specifications, using OLS to estimate the production
function is likely to generate biased estimates of factor elasticities as productivity shocks
are probably correlated with capital, labor, energy, and materials. For example, the
introduction of capital-biased technologies is likely to be associated with greater use of
capital and energy and with less employment.
We deal with this omitted variable bias by using demand-shift instruments which are
correlated with input use but uncorrelated with productivity shocks. In particular, we
construct Shea (1993) and Syverson (2003) type instruments by selecting industries whose
output fluctuations are likely to function as approximately exogenous demand shocks for
other industries. The instruments are total output measures in downstream industries, or
combinations of downstream industries, that satisfy two conditions: (1) they buy a large
enough fraction of the output generated by the upstream industry (i.e., the relevance
condition), and (2) their purchases from the output industry are a relatively small share of
their costs (i.e., the exogeneity condition). The latter condition is important to ensure that
the upstream industry is not affecting the downstream industry, so that the instrument—
downstream demand—is uncorrelated with productivity shocks upstream. The first
condition is important to ensure relevance of the instruments, i.e., strong correlations
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 345
between the instruments and use of inputs upstream. Following Shea (1993), we use the
input–output matrix for each year to construct instruments for the equivalent of two-digit
industries in the U.S.11 For the instruments to meet the two criteria above, we impose that:
(1) the demand share of the downstream industry for upstream production has to exceed
15%, and (2) the cost share of the upstream industry products in the downstream’s total
costs is less than 15%. 12 We also use one- and two-period lags of the demand shifters. The
rationale for doing this is that some factors, such as capital and labor, may face non-linear
adjustment costs and irreversibilities, so that demand shifts may take various periods to
affect factor utilization. In addition, we use regional government expenditures, excluding
government investment, in the region where the plant is located as an instrument to capture
shifts in local demand.13 We regard this instrument as particularly relevant to identify the
coefficient on labor, given that labor markets in Colombia are largely region-specific.
Finally, we use energy and materials prices as instruments, which are negatively correlated
to energy consumption and use of materials, but likely to be uncorrelated with productivity
shocks.
Table 3 reports results for the value-added specification with only capital and
employment and for the KLEM specification of the production function. Column (1)
reports the OLS results from the estimation of the value-added specification. The results
show an elasticity for capital of about 0.3 and an elasticity for employment of about 0.74,
12 While Shea (1997) uses a criterion of exogeneity based on the ratio of demand share to cost share (i.e., this
ratio being less than 3), our criterion is based on absolute cost shares (i.e., having a cost share of less than 15%).
This absolute criterion is more binding than the ratio criterion for downstream industries that represent large
demand shares, and less binding for other downstream industries. Using the above criteria, adding exports to the
downstream demand set and also aggregating downstream industries to two or three industries to meet the greater
than 15% threshold, we obtain instruments for 8 of the 12 upstream industries (i.e., Drinks and Beverages, Wood,
Food, Textiles, Oil, Paper, Chemicals and Rubber, and Glass and other non-metallic products). The 4 sectors
without demand-shift instruments are Tobacco, which makes up a very low share of sales for other industries;
Metals and Machinery, which comprise too high a share of costs when the sales share is high enough; and Other
Manufacturing, a category that includes widely different and changing types of goods, for which costs and sales
shares are difficult to define in a meaningful and consistent manner.
11 We use input–output matrices from the national accounts for the pre-1994 period. For the later years, a new
methodology was put in place for the national accounts, and input–output matrices were replaced by output–use
and output–supply matrices. For these later years, we rely on output–use matrices to determine cost and sales
shares, and on output–supply matrices to determine sectoral output. It is also important to note that input–output
matrices do not use ISIC codes to classify industries. The level at which we could create concordance is close to
the two-digit ISIC codes.
13 There may be a common component in technology shocks, although given the enormous heterogeneity
across plants the idiosyncratic component dominates. However, this common component of technology shocks is
unlikely to be correlated with regional government expenditures, since it takes time for government expenditures
to respond to changes in the local economy. Moreover, although certain types of government expenses can
constitute aggregate productivity shocks (e.g., construction of road ways), these are likely to show up in the
investment side of the government accounts. This is the reason why we use regional government expenditures,
excluding government investment, as an instrument. It is also worth emphasizing that results in the remainder of
the paper are robust changing this instrument to total regional government expenditure (which includes
government investment), and even to state GDP. That is, we obtain virtually the same results for the estimation of
the production functions and all the results that depend upon our TFP and demand shock estimates. Most results
are also robust to the exclusion of this instrument but, as we expected, the coefficient of labor hours is very
imprecisely estimated in this case, and even has the wrong sign.
Table 3
Production function equations
OLS 2SLS
(1) (2) (3) (4)
Capital 0.3226
(0.004)
0.0764
(0.0025)
0.4646
(0.0165)
0.3027
(0.0225)
Labor 0.7411
(0.0071)
1.0398
(0.0596)
Labor hours 0.2393
(0.0037)
0.2125
(0.0313)
Energy 0.124
(0.0028)
0.1757
(0.0143)
Materials 0.5891
(0.0026)
0.2752
(0.0095)
R2 0.5806 0.8621 0.4787 0.8107
N 42,235 48,114 42,235 48,114
The standard errors are reported in parentheses. The regressions in columns (1) and (3) use the log of value added
as the dependent variable, where value added is revenue minus materials and energy costs, and capital and
employment as the regressors. The regressions in columns (2) and (4) use the log of physical output as the
dependent variable, and capital, employment hours, energy, and materials as regressors. All regressors are in logs.
The two-stage least squares regressions use the following variables to instrument the inputs: Shea’s (1993)
downstream demand instruments constructed as the demand for the intermediate output (calculated using the
input–output matrix); one- and two-period lags of downstream demand; regional government expenditures,
excluding government investment; and energy and material plant-level prices, deviated from the yearly PPI.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371346
which are in line with other studies. The problem with this specification in terms of
estimating total factor productivity is, however, that it imposes implausible restrictions in
terms of the elasticity of output to energy and materials. For this reason, we estimate a
KLEM specification that uses physical output in the left-hand side and also includes
energy and materials as inputs. Furthermore, the inclusion of energy can help account for
different levels of capital utilization, under the plausible assumption that the latter is
positively correlated with the use of energy. Column (2) presents the OLS results from the
estimation of the KLEM specification.14 As expected, factor elasticities for capital and
labor become smaller, as energy and materials are allowed to impact output in a nonlinear
fashion. The results now suggest elasticities for capital, labor, energy, and materials of
about 0.08, 0.24, 0.12, and 0.59, respectively.
However, as pointed out above, these elasticities are likely to be biased if productivity
shocks are correlated with input use. Columns (3) and (4) of Table 3 present results using
2SLS estimation, which rely on the current and lagged demand-shift instrument, regional
government expenditures, and input prices. Even if we think these instruments are weakly
correlated with productivity shocks, large biases could be introduced when using IV
14 The increase in the number of observations from columns (1) to (2) (and also from columns (3) to (4)) is
due the fact that column (1) uses the log of value-added while column (2) uses the log of physical output as the
dependent variable. Observations are lost in the value added specification because the level of value added takes
negative values in some cases.
Table 4
First-stage regressions of inputs for production function equations
Instrumental variables Capital (1) Labor (2) Energy (3) Materials (4)
Downstream demand-shift 0.0157
(0.0362)
�0.1059
(0.0207)
0.2005
(0.032)
0.2506
(0.0317)
One-period lag of downstream
demand-shift
�0.1368
(0.0499)
�0.0483
(0.0286)
�0.1456
(0.0441)
�0.1619
(0.0437)
Two-period lag of downstream
demand-shift
0.3752
(0.0318)
0.0558
(0.0182)
0.3142
(0.0281)
0.3005
(0.0278)
Regional current government
expenditure
0.0484
(0.0069)
0.085
(0.004)
�0.0633
(0.0061)
�0.0166
(0.0061)
Plant-level energy price �0.2415
(0.0199)
�0.1257
(0.0114)
�0.5676
(0.0176)
�0.166
(0.0175)
Plant-level materials price �0.0253
(0.0188)
0.0386
(0.0108)
0.0166
(0.0166)
�0.8157
(0.0164)
R2 0.0319 0.0127 0.064 0.1129
Physical output partial R2 0.1276 0.139 0.231 0.324
Value-added partial R2 0.2563 0.1324 – –
N 48,114 48,114 48,114 48,114
Standard errors are in parentheses. The variables are in logs. The price variables are deviations from the yearly
PPI. Following Shea (1993), downstream demand-shift instruments are total output measures in downstream
industries that meet the following two conditions: (1) their demand share for upstream production has to exceed
15%, and (2) the share of the upstream industry in their total costs does not exceed 15%. The R2 simply reports
the square of the sample correlation coefficient between Ijt and I jt, where I=K,L,E,M and I jt are the predicted
values of the inputs from a regression of Ijt on the instruments. The partial R2 reports the sample correlation
coefficient between sjt and sjt, where sjt are the residuals from a regression of Ijt on all other inputs I1jt and sjt are
the correlations between I jt and the predicted values of all other inputs I1jt.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 347
estimation if instruments are weakly correlated with the inputs.15 To check whether inputs
are highly and significantly correlated with the instruments, Table 4 reports results for the
first-stages of the inputs on the various instruments. The results suggest use of energy and
materials increases significantly with downstream demand, but employment hours
decrease significantly with downstream demand. Moreover, capital, energy use and
materials are negatively and significantly correlated with the one-period lag of
downstream demand, and all inputs are positively and significantly correlated with the
two-period lag of downstream demand. Regional government expenditures are positively
and strongly correlated with capital and labor and negatively correlated with materials and
energy consumption. Also, as expected, materials and energy prices significantly reduce
energy consumption and use of materials and capital, respectively. Given that we are
15 We also considered longer lags and exponential terms of the demand-shifters as well as other measures of
regional shocks, including total regional government expenditures and state GDP. The results reported here are
largely robust to these changes in the list of instruments. To select our baseline instruments, we tested for
overidentifying restrictions using a variant of Basmann’s (1960) test, where we estimate a regression of the
production function residual on the instruments. While lags of more than two periods and exponential terms of the
demand shifters turn out to be significant in these regressions, the rest of our instruments are individually and
jointly insignificant. We restrict our instrument list to include those instruments that are not jointly significant in
this regression. Although regional government expenditures (excluding government investment), total regional
government expenditures, and state GDP all turn out to be insignificant in this regressions, we rely on regional
government expenditures (excluding investment) for the reasons pointed out in Footnote 13.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371348
considering instrument relevance with multiple endogenous regressors, we report the
partial R2 measures suggested by Shea (1997) for the first-stages, which capture the
correlation between an endogenous regressor and the instruments after taking away the
correlation between that particular regressor with all other endogenous regressors.16 The
partial R2’s for capital, employment hours, energy, and materials in the KLEM specification
are 0.1276, 0.139, 0.231, and 0.324, respectively, and 0.2563 and 0.1324 for capital and
employment in the value-added specification, showing that the relevant instruments for
each input can explain a substantial fraction of the variation in the use of that input.
The IV results for the value-added specification in column (3) of Table 3 show larger
capital and employment elasticities relative to the OLS estimated elasticities. The results in
column (4) of Table 3 also show larger elasticities for capital and energy, but smaller
elasticities for labor hours and materials, when inputs are instrumented.17 The results from
the value-added and KLEM specifications, thus, indicate that productivity shocks during
the period of study are positively correlated with hours and materials and negatively
correlated with capital, employment and energy, suggesting the adoption of hour-intensive
and energy-saving technologies in Colombia during the 1980s and 1990s.
4.2. Estimation of demand shocks
While productivity is likely one of the crucial components of profitability, other
components of profitability are also probably important determinants of reallocation. For
example, even if plants are highly productive, they may be forced to reduce their market
shares if faced with large negative demand shocks. We capture the demand component of
profitability by estimating establishment-level demand shocks as the residual of the
following inverse-demand equation:18
Pjt ¼ Y�ejt Djt;
where Djt is a demand shock faced by firm j at time t and �e is the inverse of the elasticityof demand and 1+e is the mark-up. We estimate this inverse-demand function in logs,
logPjt ¼ � elogYjt þ logDjt;
and our demand shock measure is estimated as the residual from this regression,
djt ¼ logDjt^ ¼ logPjt þ eelogYjt; ð2Þ
16 The standard R2 simply reports the square of the sample correlation coefficient between Ijt and I jt, where
I=K,L,E,M and I jt are the predicted values of the inputs from a regression of Ijt on the instruments. The partial R2
reports the square of the sample correlation coefficient between gjt and gjt, where gjt are the residuals from an
OLS regression of Ijt on all other inputs I1jt and gjt are the residuals from an OLS regression of Ijt on the predicted
values of all other inputs I1jt.17 The value-added specification ignores changes in labor utilization. If technology shocks are negatively
correlated with employment but positively correlated with labor hours, this would explain why the elasticity of
labor increases when the value-added specification is instrumented, while the elasticity of labor hours falls when
the KLEM specification is instrumented.18 There are other important components of profitability that we do not measure, such as access to credit
markets.
Table 5
Inverse-demand equations
Regressor OLS 2SLS First stages
(1) (2) (3) (4) (5) (6) R2
Partial R2
regression
Partial R2
regression
(5) (6)
Physical output �0.0905
(0.0011)
�0.0858
(0.0011)
�0.0447
(0.0015)
�0.4381
(0.0034)
�0.4456
(0.0035)
�0.3158
(0.0051)
0.2177 0.4696 0.427
Post-reform 0.9433
(0.0232)
2.4203
(0.0730)
– – –
Physical
output�post-reform
�0.0056
(0.0003)
�0.0909
(0.0021)
0.0104
(0.0005)
�0.2152
(0.0067)
0.9585 0.9599 0.4621
R2 0.0766 0.0793 0.0966 �1.0540 �1.0643 �0.9578 – – –
N 86,251 86,251 86,251 86,251 86,251 86,251 86,251 86,251 86,251
Standard Errors are in parentheses. The dependent variable is the plant-level price minus the yearly PPI (all in
logs). Two-stage least squares regressions instrument physical output and the interaction of physical output with
the TFP measure estimated using column (4) of Table 3 and with the post-reform dummy and the TFP measure
interacted with the post-reform dummy. The post-reform dummy takes the value of 1 for the period 1991–1998
and 0 otherwise. The R2 simply reports the square of the sample correlation coefficient between Yjt and Yjt, where
Yjt is the predicted value of output from a regression of Yjt on the instruments. The partial R2 reports the sample
correlation coefficient between ujt and ujt, where ujt are the residuals from a regression of Yjt and Yjt�Postt and
ujt are the correlations between Yjt and with the predicted value Yjt�Postt, when only shift changes are
considered, and with the interaction between the predicted value of output, Yjt, and the predicted value of the
post-reform dummy.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 349
where �e is an inverse measure of the elasticity of demand. Using OLS to estimate the
inverse-demand function is likely to generate an upwardly biased estimate of demand
elasticities because demand shocks are positively correlated to both output and prices, so
that eˆwill be smaller in absolute value than the true e. To eliminate the upward bias in our
estimates of demand elasticities, we propose using TFP as an instrument for Yjt since TFP
is positively correlated with output (by construction) but unlikely to be correlated with
demand shocks.
Table 5 reports the results of the inverse-demand equations using both OLS and IV
estimation. Columns (1)–(3) present OLS results of inverse-demand equations, some of
which impose the same elasticity while others allow different elasticities for the pre- and
post-reform periods. Columns (4)–(6) present equivalent 2SLS results. 19 Column (1),
which presents OLS results when imposing the same elasticity before and after the
reforms, suggests a large elasticity of 11.05 for the entire period of study. The results in
Column (2) which allow for different elasticities before and after the reforms suggest
demand elasticities of 11.66 and 10.94 during the pre- and post-reform periods,
respectively.20 By contrast, the results in Column (3), which allow for different intercepts
19 The sample size is larger in this table than in Table 3 because the estimations in Table 3 require information
on the instruments used for estimating the production function, while demand estimations only require
information on output prices, physical output, and TFP estimates.20 The pre-reform elasticity is calculated as the inverse of the estimated physical output coefficient. And, the
post-reform elasticity is calculated as the inverse of the sum of the estimated physical output coefficient and the
estimated coefficient of physical output interacted with the post-reform dummy.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371350
and different elasticities after the reforms, show elasticities of 22.37 and 7.37 during the
pre- and post-reform periods, but higher demand after the reforms. As indicated, however,
caution needs to be exercised in interpreting these as elasticities since these coefficients
capture the fact that higher demand implies a positive correlation between output and
prices.
The 2SLS results, which use TFP as an instrument for output, indeed show much lower
elasticities.21 The results in column (4) show a demand elasticity of 2.28, while the results
in column (5) show a demand elasticity of 2.24 during the pre-reform period and of 2.3
during the post-reform period. By contrast, allowing for different intercepts and elasticities
before and after the reforms suggests elasticities of 3.17 before the reforms and of 1.88
after the reforms, but higher demand after the reforms at given price levels.22 The last three
columns in Table 5 report high total and partial R2’s.23 The partial R2’s for physical output
and physical output interacted with the post-reform dummy on TFP, a post-reform dummy,
and TFP interacted with the post-reform dummy are of 0.427 and 0.4621, indicating
relevance of the instruments. These results suggest quantities responded less to changes in
prices but overall demand was higher after the reforms. Increased demand after reforms
may be explained by scale effects generated by trade liberalization measures and by the
downward shift in average production costs induced by factor market reforms.
5. TFP, demand, and prices before and after the reforms
In this section we rely on the TFP and demand shock measures estimated in the last two
sections as well as on output prices to document differences in the productivity and
demand components of profitability in Colombia during the 1980s and 1990s in terms of
heterogeneity and persistence. We are, for example, interested in documenting whether
productivity is as heterogeneous and persistent in Colombia as it is in other countries.
Before turning to these issues, however, we report in Table 6 basic descriptive statistics
of TFP, demand shocks and output prices for the periods before and after the reforms. The
22 Robustness analysis suggests that while it is important to use a TFP measure as an instrument in the
demand equations for output (i.e., elasticities are quite different between OLS and 2SLS in demand equation), it
matters much less whether the TFP used is from OLS or 2SLS estimation of production equation. This reduces
concerns about sensitivity of results to instrument selection for production equations. In addition, we considered
energy and materials prices as instruments for physical output in the inverse-demand equations and we find that
the results are very similar whether we use TFP alone or TFP and energy and materials prices as instruments.
21 The negative R-squares for the 2SLS are not surprising and should not be viewed as alarming. Output and
the demand shocks are highly positively correlated (this is the entire reason for using 2SLS rather than OLS).
Using the simple inverse-demand equation, the variance of prices will be equal to terms involving the variance of
output, the variance of demand shocks and a term that depends negatively on the covariance of output and
demand shocks (given that the demand elasticity is negative). Thus, the variance of demand shocks will exceed
the variance of prices and, hence, the negative R-squares.
23 The total R2 simply reports the square of the sample correlation coefficient between Yjt and Yjt, where bYjt
is the predicted value of output from a regression of Yjt on the instruments. The partial R2 reports the square of the
sample correlation coefficient between ujt and ujt, where ujt are the residuals from an OLS regressions of Yjt on
Yjt�Post-reform and ujt are the residuals from an OLS regression of the predicted value of Yjt on the predicted
value of the other regressors of the demand equation.
Table 6
Descriptive statistics of TFP, demand shocks and prices, before and after the reforms
Pre-reform Post-reform
Total factor productivity (TFP) 0.0125
(0.6817)
0.046
(0.909)
qTFP,D=0.1114 qTFP,D=�0.0995
qTFP,P=�0.5205 qTFP,P=�0.6934
qTFP,output=0.427 qTFP,output=0.506
qTFP,share=0.1297 qTFP,share=0.1836
Demand shock (D) �0.0723
(0.7826)
0.0324
(0.8836)
qD,P=0.391 qD,P=0.4995
qD,output=0.8337 qD,output=0.6243
qD,share=0.3778 qD,share=0.2863
Output price (P) �0.0764
(0.4395)
�0.1509
(0.7414)
qP,output=�0.1822 qP,output=�0.365
The table reports means, standard deviations in parentheses, and bivariate correlations. The physical output
measure of TFP is estimated as the residual of the production function in column (4) of Table 3. The demand
shock is estimated as the residual of the inverse-demand equation in column (4) of Table 5. Output prices are
plant-level prices relative to yearly PPI. Shares are estimated as the fraction of plant output in total output.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 351
table shows an increase in the simple means of total factor productivity and our demand
shock measure, but a decrease in the mean of output prices. Caution needs to be used in
interpreting these simple averages. For TFP, as we emphasize later aggregate TFP can be
decomposed into a term based upon the simple (unweighted) average and a covariance
term reflecting the relationship between market share and TFP. The latter can be
interpreted as an index of the efficiency of the allocation of resources and as we show later
is very important in Colombia. With respect to relative prices and demand shocks, note
that the average price is the average price of plant output relative to the PPI. The average
demand shock is the average of the demand shocks constructed from a micro model of
relative prices across plants. On the other hand, the table shows a substantial increase in
the dispersion of TFP, demand shocks, and output prices. The standard deviations of TFP,
demand shocks and prices increased from 0.68 to 0.91, from 0.78 to 0.88, and from 0.44 to
0.74, respectively. In addition, as expected, Table 6 shows a negative correlation between
prices and TFP and output, and a positive correlation between demand shocks and prices
both before and after the reforms, though the correlations are stronger after the reforms. By
contrast, TFP and demand shocks are positively correlated during the pre-reform period
but negatively correlated after the reforms. This correlation points to the effects of trade
reform: after trade liberalization, producers in sectors most exposed to import competition
will simultaneously increase productivity and lose demand. In contrast, manufacturers in
sectors least exposed to import competition will have relatively less incentives for
productivity enhancement and will less likely face a drop in demand. In addition, the table
shows that firms facing higher demand and productivity shocks produce more output, both
before and after reforms. However, while the relation between output and productivity
becomes stronger after the reforms, the relation between output and demand becomes
weaker. This correlation suggests, again, that the reforms were successful in strengthening
Fig. 1. TFP dispersion for continuing, entering, and exiting plants.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371352
the link between a business’s performance and its productivity. Moreover, this table shows
that not only do more productive firms increase production after the reforms, but they also
increase their share in the market. This points to the increased importance of productivity-
enhancing reallocation after the reforms, which we document in Section 6.
5.1. Heterogeneity
Previous evidence for the US suggests a pattern of evolving common productivity and
prices among heterogeneous plants (Baily et al., 1992; Foster et al., 2001; Foster et al.,
2003; Roberts and Supina, 1996). Like in the U.S., dispersion measures in Table 6 suggest
much heterogeneity in productivity and prices across Colombian plants. Figs. 1–3 show
the evolution of dispersion measures of TFP, demand shocks and output prices over the
entire period of study.24 The figures show large percentage increases in the dispersion of
productivity and prices over the last two decades. In particular, Figs. 1 and 3 shows large
rises in the dispersion of productivity and prices of entering and exiting plants after 1990,
and a constant increase in the dispersion of productivity and output prices of continuing
plants throughout the entire period. By contrast, Fig. 2 shows a much smaller percentage
increase in the dispersion of demand over the last two decades. The greater heterogeneity
in productivity is consistent with greater process experimentation across businesses after
24 We have examined these patterns using alternative robust measures of dispersion (in particular, the 90–10
differential) and have found very similar patterns overall for TFP, prices and demand shocks.
Fig. 2. Dispersion of demand shocks for continuing, entering, and exiting plants.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 353
reforms, while the modest increase in the dispersion of demand is consistent increased
competition, and also with modest increased experimentation in product variety across
businesses.
Fig. 3. Output prices dispersion for continuing, entering, and exiting plants.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371354
Figs. 4–6 show Kernel distributions of TFP, demand shocks and output prices for the
pre- and post-reforms periods, which illustrate changes in higher moments of the
distributions. The distributions display fatter tails (i.e., increased kurtosis) of the
distributions of TFP, demand shocks and output prices after the reforms, suggesting that
more firms were experiencing either positive or negative large shocks after the reforms.
5.2. Persistence
To examine whether Colombian plants, like U.S. plants, are characterized by a large
degree of persistence, we estimate AR(1) models for productivity, demand shocks and
prices. Columns (1), (3) and (5) in Table 7 report the coefficients on the lags of
productivity, demand and output prices for AR(1) models with year effects. Our estimates
show a great deal of persistence of productivity, demand shocks and prices in Colombia,
much greater than that found for the U.S. The results suggest that 67% of all plants
continue with their initial productivity after 5 years, 93% of plants face the same demand
shocks after 5 years, and 96% charge the same prices as 5 years ago. Studies for the U.S.
find much lower degrees of persistence. For example, Bartelsman and Dhrymes (1998)
find that after 5 years only about a third of all plants remain in the same productivity
quintile. Foster et al. (2003) also find that about a third of plants remain in their original
position in the productivity and price distributions after 5 years. In spite of the reforms
increased competition after the reforms, the greater persistence in the Colombian context
probably reflects continued presence of distortions in this economy.
Fig. 4. Kernel distribution of TFP.
Fig. 6. Kernel distribution of relative prices.
Fig. 5. Kernel distribution of demand shocks.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 355
Table 7
Persistence of TFP, demand shocks, and output prices
Regressor TFP Demand shocks Output prices
(1) (2) (3) (4) (5) (6)
One-period lag of
dependent variable
0.9221
(0.0018)
0.896
(0.0029)
0.9853
(0.0009)
0.9871
(0.0014)
0.9925
(0.0013)
0.982
(0.0026)
One-period lag of
dependent variable�postQreform
0.043
(0.0037)
�0.0033
(0.0019)
0.0144
(0.003)
Year effects YES YES YES YES YES YES
R2 0.782 0.7824 0.9305 0.9305 0.868 0.868
N 73,063 73,063 85,576 85,576 85,576 85,576
Standard errors are in parentheses. The table reports coefficients on lagged TFP, lagged demand shocks, and
lagged output prices in the first and second, third and fourth, and fifth and sixth columns, respectively. All
regressions include year effects. The TFP variable is estimated as the residual from the regression in column (4) of
Table 3 and the demand shock variable is estimated as the residual from the regression in column (4) of Table 5.
Output prices are plant-level prices relative to yearly PPI.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371356
Columns (2), (4), and (6) in Table 7 also report how the degree of persistence changes
after the market reforms. The results suggest persistence in productivity and output prices
increased after the reforms, while the persistence of demand shocks did not change. At
first glance, this pattern might be surprising as one might conjecture that market reforms
would lead to greater mobility in the distributions of efficiency, demand shocks, and
prices. However, it is important to emphasize that these persistence measures are
conditional on survival. Our results are consistent with increased weeding out of less
productive plants, as reported in the next section, but lower mobility of the plants that do
survive after the reforms.
6. Productivity- and profitability-enhancing reallocation
In this section, we examine whether the market reforms changed the allocation of
activity in terms of productivity and demand factors. In particular, we are interested in
whether increased competition as a result of the reform process induced reallocation from
less to more productive plants and reduced the role of demand factors in the allocation of
activity. Similarly, we are interested in whether increased competition as a result of the
reforms increased productivity by making it more difficult for less productive plants to enter
the market and survive. Different types of models have been proposed which can explain
productivity-enhancing reallocation of this sort. In learning models (e.g., Jovanovic, 1982;
Ericson and Pakes, 1995), experimentation generates differences in productivity and
successful experimenters gain market share over unsuccessful ones. Similarly, vintage
capital models can explain gains in market share of plants with newer vintages over those
with older vintages. If more efficient plants can afford more frequent retooling, initial
productivity differences would be exacerbated. Moreover, if trade liberalization and FDI
increased competition in product markets after the reform process, this may have speeded
up learning and productivity-enhancing reallocation. Similarly, if financial liberalization in
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 357
1992 improved access to credit markets, this may have provided access to better capital
vintages and increased productivity-enhancing reallocation after the reforms.
6.1. Productivity growth and reallocation
We use a cross-sectional decomposition methodology, first introduced by Olley and
Pakes (1996), to examine whether the allocation of activity in the economy has become
more or less productivity-enhancing over our period of study. We conduct the following
decomposition of aggregate TFP and total demand shocks, constructed from our estimates
of plant-level TFP and demand:
TFPt ¼PTFPt þ
XJj¼1
ð fjt � f tÞ TFPjt �PTFPt
�;
�
dt ¼ d t þXJj¼1
fjt � f t�djt � d t
���
where TFPt and dt are aggregate total factor productivity and demand shock measures for
a given three-digit manufacturing sector in year t. These aggregate measures correspond to
weighted averages of our plant-level measures TFP and demand shocks, where the weights
are market shares (calculated as described below).PTFPtand dt are the cross-sectional
(unweighted) means of total factor productivity and demand shock measures across all
plants in that sector in year t, TFPjt and djt are total factor productivity and demand shock
measures of plant j at time t estimated as described in Section 4, fjt is the share or fraction
of plant j’s output out of sectoral output at the three-digit level in year t, and ft is the cross-
sectional unweighted mean of fjt.25 The second term in this decomposition allows us to
understand whether production is disproportionately located at high-productivity or high-
demand plants. Moreover, by examining this decomposition over time we can learn
whether the cross-sectional allocation of activity has changed in response to the market
reforms.26
Table 8 shows the cross-sectional decompositions for TFP and demand shock measures
from 1982 to 1998. We report the results for the average three-digit sector. Column (1)
shows a positive trend in aggregate total factor productivity over the period of study.
25 This means that our focus here is on within sector allocation and reallocation rather than between sector
allocation, for sectors defined at the three-digit level. For measurement and conceptual reasons, comparisons of
TFP and relative demand across sectors (in levels) are more problematic to interpret. Focusing on within sector
allocation permits us to emphasize the degree to which market reforms have led to an improved allocation of
activity across businesses due to higher competition. Nevertheless, we have also tried a decomposition using the
shares or fraction of plant output out of total output. The results based on this alternative decomposition are
similar to those reported below.26 An advantage of this cross-sectional method over methods that decompose changes in productivity over
time is that cross-sectional differences in productivity are more persistent and less dominated by measurement
error or transitory shocks. Another advantage is that this method does not rely on accurate measurement of entry
and exit. The downside being, of course, that this decomposition does not allow characterizing the role of entry
and exit. We explore the latter in Section 6.2.
Table 8
Cross-section decomposition of three-digit level TFP and demand shocks, 1982–1998
Year TFP Demand shocks
Aggregate
(Weighted) (1)
Simple
Average (2)
Cross-term
(3)
Aggregate
(Weighted) (4)
Simple
Average (5)
Cross-term
(6)
1982 0.28
(100)
�0.05
(�16.46)
0.33
(116.46)
1.26
(100)
0.04
(3.30)
1.22
(96.70)
1983 0.29
(100)
�0.09
(�30.80)
0.38
(130.80)
1.22
(100)
0.03
(2.12)
1.20
(97.88)
1984 0.26
(100)
�0.07
(�28.57)
0.33
(128.57)
1.27
(100)
0.08
(6.31)
1.19
(93.69)
1985 0.33
(100)
0.00
(�0.28)
0.33
(100.28)
1.27
(100)
0.07
(5.73)
1.20
(94.27)
1986 0.33
(100)
0.06
(16.49)
0.28
(83.51)
1.27
(100)
0.07
(5.76)
1.19
(94.24)
1987 0.36
(100)
0.07
(18.99)
0.29
(81.01)
1.30
(100)
0.09
(6.73)
1.22
(93.27)
1988 0.42
(100)
0.14
(32.21)
0.29
(67.79)
1.32
(100)
0.12
(8.86)
1.20
(91.14)
1989 0.40
(100)
0.11
(28.71)
0.29
(71.29)
1.28
(100)
0.08
(6.23)
1.20
(93.77)
1990 0.45
(100)
0.10
(22.82)
0.35
(77.18)
1.28
(100)
0.10
(7.63)
1.18
(92.37)
1991 0.48
(100)
0.12
(24.40)
0.36
(75.60)
1.24
(100)
0.07
(6.04)
1.16
(93.96)
1992 0.46
(100)
0.05
(10.59)
0.41
(89.41)
1.26
(100)
0.17
(13.62)
1.09
(86.38)
1993 0.43
(100)
0.05
(11.99)
0.38
88.01)
1.28
(100)
0.23
(17.74)
1.06
(82.26)
1994 0.47
(100)
0.03
(6.55)
0.44
(93.45)
1.28
(100)
0.24
(18.35)
1.05
(81.65)
1995 0.46
(100)
�0.04
(�8.44)
0.50
(108.44)
1.28
(100)
0.21
(16.74)
1.07
(83.26)
1996 0.53
(100)
�0.05
(�8.77)
0.58
(108.77)
1.21
(100)
0.17
(14.15)
1.04
(85.85)
1997 0.54
(100)
0.01
(2.71)
0.53
(97.29)
1.21
(100)
0.16
(13.31)
1.05
(86.69)
1998 0.58
(100)
0.04
(6.95)
0.54
(93.05)
1.13
(100)
0.11
(9.52)
1.03
(90.48)
Pre-reforms 0.35
(100)
0.03
(8.56)
0.32
(91.44)
1.27
(100)
0.08
(5.88)
1.20
(94.12)
Post-reforms 0.50
(100)
0.03
(5.50)
0.47
(94.50)
1.24
(100)
0.17
(13.78)
1.07
(86.22)
All figures are simple means of three-digit sector level statistics. Columns (1) and (4) show the weighted mean of
TFP and the demand shocks, respectively, where market shares are the weights. The market shares are measured
as the share or fraction of output contributed by each plant to sectoral output defined at the three-digit level.
Columns (2) and (5) show the contribution to those weighted means of the simple means of TFP and the demand
shock, respectively. Columns (3) and (6) show the contribution of the cross-sectional correlation between market
share and TFP and demand shocks, respectively. The numbers in parentheses are the fractions represented by each
component.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371358
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 359
Columns (2) and (3) show the decomposition of aggregate productivity into the simple
average and the cross-sectional correlation between market share and total factor
productivity. In all years, aggregate productivity is largely accounted for by the allocation
of activity to higher productivity businesses, as shown by the large fraction that the cross-
term represents out of aggregate productivity. This stands in sharp contrast to results of
similar exercises for the U.S., where the contribution of this term is only marginal (see
Foster et al., 2001). In a context in which the production possibilities frontier is not
improving sharply over time, the allocation of activity seems to be the crucial factor
accounting for the level of aggregate productivity.
The cross-sectional correlation between market share and total factor productivity
remained roughly constant over the 1980s, with a slight decrease by the middle of the
decade. By contrast, it increased sharply during the 1990s, most specially after 1993, once
most reforms were fully implemented. This time-series pattern can be seen in Fig. 7 and
suggests that market share increasingly shifted towards more productive plants and away
from less productive plants during the 1990s. In fact, while average productivity explains
about 8.6% of aggregate productivity during the pre-reform period, it only explains about
5.5% of aggregate productivity during the post-reform period.
Columns (1)–(3) in Table 9 provide evidence of the statistical significance of the effects
of reforms on the Olley–Pakes decomposition of TFP, just described. Each of these
columns reports results of regressing one of the terms of the TFP decomposition on a post-
reform dummy that takes a value of 1 in 1991–1998, and ISIC three-digit sector controls,
using data at the three-digit level. Column (1) shows results using aggregate TFP (the
weighted average of TFP using market shares as weights) as dependent variable. The
Fig. 7. Cross-sectional decomposition of TFP, 1982–1998.
Table 9
Decomposition of conditional means of TFP and demand shocks, before and after reforms
Regressor TFP Demand shocks
Aggregate
(weighted)
(1)
Simple
average
(2)
Cross-term
(3)
Aggregate
(weighted)
(4)
Simple
average
(5)
Cross-term
(6)
Post-reform
dummy
0.1472
(0.0231)
�0.0026
(0.0171)
0.1498
(0.0185)
�0.0362
(0.0152)
0.0957
(0.0140)
�0.1319
(0.0131)
Sector
effects
YES YES YES YES YES YES
R2 0.7022 0.6552 0.5939 0.926 0.9078 0.8456
N 493 493 493 493 493 493
Standard errors are in parentheses. The table reports coefficients of a regression of one term of TFP or demand
decomposition on a constant and a post-reform dummy. The regression is at three-digit level. The post-reform
dummy takes the value of 1 in 1991–1998, 0 otherwise. In columns (1) and (4) the dependent variable is the
weighted mean of TFP and the demand shocks, respectively, where market shares are the weights. In columns (2)
and (5) the dependent variable is the simple mean of TFP and demand shocks, respectively. In columns (3) and (6)
the dependent variable is the cross-sectional correlation between market share and TFP and demand shocks,
respectively.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371360
results show a statistically significant increase in aggregate TFP after the reforms. Column
(2) reports results of a regression of the simple average of TFP. These results show no
significant change in the unweighted mean of TFP. Finally, column (3) reports results of a
regression of the cross-sectional correlation between market share and TFP. The results
show a significant increase of almost 50% (with respect to the magnitudes reported in
Table 8) in the correlation between market share and TFP after the reforms. Thus, the post-
reform period is characterized by a significant increase in the ability of highly productive
firms to gain market share.27
The last three columns of Table 8 show the Olley–Pakes decomposition of our demand
shock measure, since we are also interested in whether the reform process changed the role
of demand factors in the allocation of activity. Column (4) shows a decreasing trend of
aggregate demand shocks over the period of study, although the average plant actually
perceived an increase in demand (column (5)). Moreover, in contrast to the results on
productivity, column (6) suggests that the concentration of activity in high demand plants
became less important during the 1990s. Fig. 8 shows the time-series pattern of this
decomposition.
The statistical significance of the effects of reforms on the decomposition of demand
shocks is analyzed in columns (4)–(6) of Table 9. As was the case for the decomposition
of TFP, each of these columns presents the results of a regression of one of the
components of the decomposition of demand on the post-reform dummy and sector
controls, using data at the three-digit sector level. The dependent variable in column (4)
is the aggregate demand shock, in column (5) it is the simple average of plant-level
27 As reform does not significantly change the simple average of TFP, rather than plant turnover what
explains the changes in aggregate TFP is the reallocation from less productive to more productive incumbents.
Indeed, below we document that on net the contribution of entry and exit to productivity rises only slightly post
reform.
Fig. 8. Cross-sectional decomposition of demand shock, 1982–1998.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 361
demand shocks, and in column (6) it is the cross-term of the decomposition. The results
indicate a negative and statistically significant change in aggregate demand shocks after
the reforms (column (4)), although the magnitude of the change is small relative to the
magnitudes of aggregate demand shocks, as reported in Table 8. Nevertheless, the
demand shocks faced by the average plant experience a positive and significant change
with the reforms. The decline in aggregate demand shocks is explained by a decrease in
the correlation between demand shocks and market share, which fell significantly after
the reforms. These effects point to a decline in the relative importance of high demand
in gaining market share.
Our results suggest that the packet of reforms, most likely through its components
of trade liberalization, removal of FDI restrictions and liberalization of factor
markets, contributed to raise productivity by improving the allocation of activity with
higher market shares at high-productivity plants and lower market shares at low-
productivity plants. At the same time, we find that, after the reforms, the allocation
of activity was driven less by high demand plants commanding larger shares. This
latter finding is consistent with the hypothesis that increased competition reduced the
influence of demand factors (i.e., markets became more competitive) in the allocation
of activity.
6.2. Entry and exit effects
Vintage capital models of industry dynamics suggest new firms are relatively more
productive, as they receive the productivity associated with the latest vintage
Table 10
Evolution of TFP, demand shocks, and output prices
Regressor TFP Demand Shocks Output Prices
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Entry 0.0288
(0.0115)
0.0592
(0.0148)
0.0847
(0.014)
0.121
(0.0178)
�0.4354
(0.0094)
�0.4441
(0.0123)
�0.4681
(0.0111)
�0.4808
(0.0143)
0.0257
(0.007)
0.0119
(0.0091)
0.0386
(0.0084)
0.0319
(0.0108)
Entry�post-ref. �0.0771
(0.0235)
�0.0968
(0.0288)
0.021
(0.0191)
0.0341
(0.0228)
0.0339
(0.0141)
0.0192
(0.0173)
1-prd. lag of entry 0.0938
(0.0133)
0.0986
(0.0157)
�0.3941
(0.0123)
�0.4002
(0.0147)
0.0242
(0.0093)
0.0265
(0.0111)
1-prd. lag of
entry�post-ref.
�0.0201
(0.0296)
0.0212
(0.0269)
�0.0058
(0.0204)
2-prd. lag of entry 0.0644
(0.014)
0.047
(0.0165)
�0.3484
(0.013)
�0.3696
(0.0155)
0.0138
(0.0099)
0.008
(0.0118)
2-prd. lag of
entry�post-ref.
0.0617
(0.031)
0.0731
(0.0284)
0.0217
(0.0216)
Exit �0.2497
(0.0102)
�0.2262
(0.0141)
�0.2666
(0.0122)
�0.2408
(0.0172)
�0.4664
(0.0094)
�0.4589
(0.0132)
�0.4829
(0.011)
�0.4885
(0.0156)
0.0306
(0.0069)
0.0087
(0.0098)
0.0319
(0.0083)
0.0018
(0.0118)
Exit�post-reforms �0.0485
(0.0204)
�0.0519
(0.0244)
�0.0152
(0.0188)
0.0117
(0.022)
0.0438
(0.0139)
0.0608
(0.0167)
1-prd. lead of exit �0.1826
(0.0131)
�0.1339
(0.0173)
�0.4028
(0.012)
�0.4675
(0.0157)
0.0184
(0.0091)
�0.0442
(0.0119)
1-prd. lead of
exit�post-ref.
�0.1144
(0.0266)
0.1534
(0.0242)
0.1481
(0.0183)
2-prd. lead of exit �0.1321
(0.0136)
�0.129
(0.0179)
�0.3371
(0.0125)
�0.3589
(0.0165)
0.022
(0.0095)
0.0131
(0.0125)
2-prd. lead of
exit�post-ref.
�0.0077
(0.0275)
0.0516
(0.0252)
0.021
(0.0191)
Year effects YES YES YES YES YES YES YES YES YES YES YES YES
R2 0.0151 0.0153 0.0162 0.0169 0.0610 0.0610 0.1078 0.1086 0.0036 0.0038 0.0026 0.0038
N 76,261 76,261 56,025 56,025 88,690 88,690 65,398 65,398 88,690 88,690 65,398 65,398
M.Esla
vaet
al./JournalofDevelo
pmentEconomics
75(2004)333–371
362
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 363
technology, and once they have entered the market their productivity remains constant
over time unless hit by random shocks. Then, firms exit the market when the
productivity relative to new entrants is below a certain threshold. However, aside from
Foster et al. (2003) who find higher productivity of entering plants and lower
productivity of exiting plants, the previous literature documents lower productivity of
both entering and exiting relative to incumbent plants (see Bartelsman and Doms,
2000), a finding that is at odds with the vintage capital model. Using our estimates of
TFP and demand shocks, we analyze this issue for the Colombian economy, and
explore whether the market-oriented reforms changed how productive entering and
exiting businesses are relative to incumbents.
Table 10 reports regressions of TFP, demand shocks, and prices on year dummies
as well as on entry and exit dummies. The entry dummy takes a value of 1 for
plant j in period t if the plant produced in year t but not in t�1. Conversely, the
exit dummy takes a value of 1 for plant j in period t if plant j produced in period
t but not in t+1. Column (1) in Table 10 shows that, consistent with the vintage
capital hypothesis, productivity of entering plants is above and productivity of
exiting plants below that of the average incumbent. Our results coincide with those
in Foster et al. (2003). Like theirs, our TFP estimation uses physical output rather
than revenue, which avoids confounding price and productivity variations in
measures of TFP. The results in column (5) also show lower idiosyncratic demand
for entering and exiting plants relative to the average incumbent, and the results in
column (9) show entrants and exiting plants charging higher prices than incumbents.
While higher prices at exiting plants are consistent with their lower productivity,
higher prices at entering plants might reflect their smaller scale or perhaps higher
markups.
Since the reforms introduced during the 1990s are generally expected to increase
competition, we also explore whether the productivity, demand and prices of entering
and exiting plants relative to incumbents changed after the implementation of these
reforms by interacting the entry and exit dummies with a post-reform dummy which
takes the value of 1 for the period 1990–1998. The results in column (2) of Table 10
show a much smaller differential entry effect on productivity after the reforms. That is,
after the reforms, entering businesses had a smaller productivity advantage relative to
incumbents in the year of entry. Lower adjustment costs due to less distorted factor
markets appear to show a greater ability of surviving incumbents to catch-up with
Notes to Table 10:
Standard errors are in parentheses. The TFP variable is estimated as the residual from the regression in
column (4) of Table 3 and the demand shock variable is estimated as the residual from the regression in
column (4) of Table 5. Entry is a dummy which takes the value of 1 if the plant entered that period and 0
otherwise, and the one- and two-period lags are dummies that take the value of 1 if the plant entered last
period or two periods ago and 0 otherwise. A plant entered in period t if it reported positive production in
period t but not in period t�1. Exit is a dummy which takes the value of 1 if the plant exits this period and
0 otherwise, and the one- and two-period leads are dummies that take the value of 1 if the plant exits one or
two periods ahead. A plant exited in period t if it reported positive production in period t but not in period
t+1.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371364
potentially highly productive entrants. First, they can retool more frequently to install
latest vintage technologies. Second, this is consistent with increased layoffs of
unproductive workers after the reduction in dismissal costs in 1990. Alternatively,
there may have been increased market experimentation after the reforms so that the
average entrant had lower initial productivity. This latter idea we explore further
below.
On the other hand, the results suggest that productivity of exiting plants relative to
incumbents fell substantially after the reforms. This pattern is consistent with increased
exit of plants which may had been kept afloat by subsidized credit, relatively low
taxes, and lack of international competition before 1990. On net, the contribution of
entry and exit to overall productivity rises slightly after the reforms. Moreover, the
results show that relative demand for entering and exiting plants did not change after
the reforms. On the other hand, relative prices of both entering and exiting plants
increased after the reforms.
We also examine learning effects by looking at the evolution of productivity,
demand, and prices of plants over the first years they are in the market. The results in
column (3) of Table 10, for the entire period, include one- and two-period lags of the
entry dummies and one- and two-period leads of exit dummies, capturing plants that
entered in t�1 and t�2 and plants that will exit in t+1 and t+2. The results show
little evidence of learning prior to market reforms, as businesses do not appear to be
more productive 1 or 2 years after entering than they were upon entrance. However,
after market reforms, there is evidence of substantial learning amongst entrants
(column (4)). As noted, the productivity advantage of entrants diminished after
reforms, but the productivity advantage 2 years after entry increased substantially. This
pattern is consistent with the hypothesis mentioned above of increased market
experimentation by entrants after the reforms.
Finally, we examine a number of other patterns in terms of entry and exit. It is
interesting to note that there is evidence of a bshadow of deathQ effect: the negative
productivity differential faced by exiting plants becomes stronger the closer to exiting
the plant is. This effect grows stronger after reforms. We also find some evidence of
QlearningQ in terms of demand and prices. In this context, we mean that entering
plants appear to gain a greater demand/price advantage as time after entry grows.
Column (7) and (11) show that, as time elapses since plant entry, the plant’s relative
demand grows and the prices it charges for its output fall. Columns (8) and (12)
show greater learning effects in demand and prices after the reforms.
In short, the results suggest that market reforms yielded a more heterogeneous
group of entrants, but that conditional on survival the post-reform entrants contribute
more to productivity. Subsidized credit prior to the 1990 and 1991 financial reforms
may have previously encouraged entry of unproductive plants and kept existing
unproductive plants afloat. Moreover, greater learning after reforms may be due to
greater access to new vintages of imported capital and greater access to the know-how
of foreign firms in the economy. In terms of exiting behavior, intensified foreign
competition after the trade reforms seems to have increased market discipline by
forcing firms either to increase productivity and charge lower prices, or else exit the
market.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 365
7. Conclusion
In this paper, we examine how reallocation contributes to productivity and other
determinants of profitability in Colombia. Then, we ask how the relation between
reallocation and productivity and demand factors changed after market reforms were
introduced in Colombia during the early 1990s. The extent, breadth and swiftness of
the reforms make Colombia a superb country to study these issues. In addition, a
unique feature of the Colombian data is that it allows us to measure both plant-level
quantities and prices, making these data ideal for measuring plant-level productivity
and demand.
We propose a sequential methodology for estimating both productivity and demand
shocks at the plant level. First, we generate measures of plant-level total factor
productivity by estimating KLEM production functions using plant-level physical output
data. To eliminate biases from the correlation between productivity shocks and inputs, we
use downstream demand shifts and plant-level energy and materials prices as instruments
in estimating production functions. We then generate plant-level demand shocks by
estimating inverse-demand equations using plant-level output prices, where output is
instrumented with TFP to eliminate biases from the correlation between demand shocks
and output.
We find some interesting patterns with regards to total factor productivity, prices
and demand shocks in Colombia, which sometimes contrast with the patterns found for
the U.S. First, as in the U.S., we find a great deal of heterogeneity in terms of
productivity and output prices, and we find that this heterogeneity has been increasing
over the past decades. Second, we find a large degree of persistence in productivity,
prices and demand, which doubles or triples the degree of persistence found in the
U.S.
We also explore turnover dynamics in Colombia during the entire period of study as
well as before and after the introduction of reforms. First, consistent with recent
evidence for the U.S. which uses plant-level prices, we find that entering business are
more productive than incumbents and the exiting businesses they are replacing, and that
exiting businesses are much less productive than incumbents. Note that revenue-based
TFP measures spuriously include a demand component and thus underestimate the
technical efficiency of entrants, who have to build a consumer base, relative to
incumbents, who have a established clientele. For this reason, our analysis provides a
better basis to discern between vintage and learning models. Our results are more
consistent with a vintage capital explanation than a learning explanation of industry
dynamics. In addition, we find that turnover is affected by demand-side factors, as both
entering and exiting plants face lower demand for their products than incumbents.
Second, we explore how these patterns changed in response to market reforms. After the
reforms, entering businesses no longer had as much initial productivity advantage
relative to incumbents, but the productivity of exiting plants relative to incumbents falls
substantially. On net, the contribution of entry and exit rises slightly. This is consistent
with increased redundancy of unproductive workers, who were previously retained by
plants due to high dismissal costs, and with increased exit of plants which may had been
kept afloat by subsidized credit, relatively low taxes, and limited foreign competition
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371366
before 1990. When we explore a more dynamic specification, we find evidence of
learning effects for surviving entrants after the reforms. This is consistent with a
contribution to learning due to trade and foreign direct investment liberalization. The
increased competition also imposed market discipline, forcing plants to either increase
productivity and charge lower prices, or exit the market. On the other hand, access to
new vintages of imported capital also contributes to productivity growth if lower capital
adjustment costs increase the optimal frequency of retooling to install equipment
embedding latest technology.
Not surprisingly, we find that the allocation of activity across businesses reflects both
efficiency and demand factors. Businesses with higher TFP and higher demand have a
higher market share. The more interesting question is how these patterns changed in
response to market reforms. We find that, after the reforms, there is a significant increase
in the importance of efficiency and a decline in the importance of demand factors in
determining productivity. This change in the efficiency of allocation has a large impact on
aggregate productivity. Specifically, we find that the higher aggregate productivity in
Colombia over this period (for the average three-digit industry) is largely accounted for by
the improved allocation of activity to high-productivity businesses. We also find that
demand factors become less important after market reforms in determining the allocation
of activity, which is consistent with the implications of increased product market
competition.
Our paper takes a first look at the impact of the reforms on the relation between
efficiency and demand-side factors and the allocation of activity by essentially asking
whether the covariance structure of productivity, demand shocks and the allocation of
activity changes after the reforms. While instructive, much research remains to be
done in this context, in particular to investigate the reforms with more information
and structure. In future work, we plan to explore the temporal and cross-sectional
variation of the various reforms more fully. Not all the market reforms happened at
the exact same time, and many of the market reforms likely impacted firms
differentially depending upon sector and other characteristics (e.g., factor intensities,
size, and location) of the businesses. Moreover, we plan to explore the dynamic
implications of the reforms in terms of their impact on market selection in
Colombia.
Acknowledgements
We thank Alberto Serrano, Juan Manuel Contreras and Pablo Medina for excellent
research assistance, Alvaro Suarez and Jose Eduardo Granados at DANE for providing
access to data, the staff in charge of the AMS for answering our queries on
methodological issues regarding the data, and participants at various seminars at DANE
for their comments and suggestions. We thank the participants of the IASE conference
in Chile in November 2003 and, especially, Andrew Bernard, Raphael Bergoeing,
Jonathan Eaton, Sebastian Edwards, Mark Roberts and Ernesto Schargrodsky for
excellent comments. Support for this project was provided by the World Bank. John
Haltiwanger also acknowledges financial support from the NSF and the IDB, Adriana
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 367
Kugler from the Spanish Ministry of Science and Technology through grant No. SEC-
2003-04429, and Maurice Kugler from the IDB and the Stanford Centre for
International Development.
Appendix A. Longitudinal linkages in the AMS data
The identification of establishments in the AMS for the period 1982–1998 has
overcome changes in sampling and labeling of plants. In essence, there are three groups
of plant identifiers: a system that allows following plants from 1982 to 1991,28 another
system that can be used to follow plants from 1993 onwards, and finally a transition
system that is supposed to provide the necessary link over the transition period
(1991–1993). Nevertheless, the latter set is quite incomplete, so that many plants that
existed prior to 1992 and survived until after 1993 cannot be successfully identified
as continuers. As a result, matching plants using the transition identifiers leads to
spurious calculations of entry and exit. To solve this problem, we relied on some
existing records of additional identifiers that, although incomplete, provide some
information about the different identifiers a given plant was assigned over the
period. This additional effort eliminated a large fraction of the spurious entry and
exit. For instance, after performing this additional matching, we are unable to follow
only 740 plants after 1991, while in previous independent attempts documented by
DANE the corresponding figure was 4083. However, 740 bdeathsQ in a year is still
too high a figure compared to the historic yearly exit rates. Similarly, we obtain too
high an entry rate for 1992 relative to historic levels, to be fully explained by
fundamentals.
Spuriously high exit and entry rates stem from two sources. First, changes in the sample
resulting from the change of survey unit from establishment to enterprise. In particular,
from 1992 if a firm that owns various establishment satisfies the requirements to enter the
sample, each of its establishments is included, independently of whether the plant itself
satisfies the requirements or not.29 Before 1992, meanwhile, only establishments that
satisfied the requirements were included in the sample, independently of their ownership.
28 This is the time period covered by the subset of AMS data used by Roberts and Tybout (1996) and Clerides
et al. (1998). Note that their sample does not include any years after the implementation of structural reforms in
Colombia, nor does it have plant-level price indices.29 These requirements are that an establishment must have 10 or more employees or have a production
superior to a given limit. Such limit was 25 million Colombian Pesos (in current terms) for 1982–1991, 50 million
for 1992–1995, 60 million for 1996–1997, and 70.5 million for 1998. One obvious problem of these requirements
is that the minimum production has always been established in current pesos, and has not been properly modified
to reflect inflation. With an average PPI inflation of around 20%, the minimum real production required for
inclusion in the sample decreased dramatically over those years where the limit was kept constant. This may be
affecting the levels of entry and exit; our measure of entry will actually be an upper bound of entry, while the
opposite will happen with our measure of exit. Notwithstanding this problem, we decided not to change the
requirements to enter the sample because the only non-arbitrary method we found (deflating the limits by some
known deflator, like PPI) leads to an underestimation of the growth of the industry. Moreover, the limit on the
number of employees is the binding requirement in most cases.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371368
To solve the problem of sample change, we deleted from our base of each year the
establishments that do not satisfy the requirements of the sample.30
Secondly, there are possible mistakes in the assignment of codes, which make
impossible to match an establishment between 1991 and 1992, even though the
establishment continues over the period. The solution to this problem implies matching
what now is a bdeath of 1991Q with what now appears to be an bentry of 1992Q. Thismatching must use additional information, such as the one contained in the directories
where data on the name, tax identification number, location, and other characteristics of
the plant are recorded. We therefore resorted to these directories. Due to differences in the
format of the information, only the data on phone number, sector identification code and
municipality were used for this match. Unfortunately, this new attempt yielded very few
matches.
Given the failure of the efforts documented above to fully solve the problem of spurious
entry and exit over the transition period, we implemented a probabilistic approach to
eliminate the remaining problems. Our methodology involves estimating the probabilities,
on the basis of observable characteristics, of entering and exiting of the plants that, after
the previous procedures, appear as exiting in 1991 or entering in 1992. We used a Probit
estimation procedure to model the probability of entry and the probability of exit in any
given year. We defined the discrete variables ENTER and EXIT for entry and exit,
respectively, with the values of 1 for event and 0 for nonevent. In our model, a plant of a
given sector and a given geographic region enters (exits) when its expected profits are
higher (lower) than an exogeneously given limit. Profits pjt are a linear function of (real)
output and number of employees. The probability of entry and exit within a sector and a
geographic region is, therefore, a function of output and employment.
Let s be a subscript for two digits sector and r be a subscript for region (Atlantic,
Pacific, Central, and Antioquia plus Viejo Caldas). Then, for plant j that belongs to sector s
and region r, let ejt=1 if plant j enters in year t, with ejt=0 otherwise, and xjt=1 if plant j
exits in year t, with xjt=0 otherwise.
The entry probability may be expressed as,
Pr ejt ¼ 1�¼ Pr pjtNp4
�¼ Pr b0; sr þ b1; srYjt þ b2; srLjt þ jtNp4
����
Assuming ejt follows a cumulative normal distribution denoted by F,
Pr ejt ¼ 1�¼ 1� F � XjtVw
� �¼ F XjtVw
� ��
where the last step follows from the fact that F is a symmetric distribution, and Xjt=[1V,YjtVx, LjtV] and wV=[b0�p*, b1, b2].
When estimating the model for entry in 1992 (exit in 1991), we used information by
sector and geographic region, where all years and plants where pooled. For each
30 One point to note is that, apparently, the change in methodology also included surveying establishments
that were not in the sample even though they did satisfy the requirements (due to previous failures in enforcing the
requirement to report to the AMS). This modification is not formally documented, but DANE staff have reported
that an effort was made in 1992 to improve the actual coverage of the survey. To the extent that this is the case, the
solution proposed above to the problem of change in sample (namely, to keep only establishments that satisfy the
sample criteria) will not fully alleviate the problem.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 369
estimation we excluded the plants for which we wanted to predict the probability of event
(plants that after the previous procedures appear as entering in 1992 or exiting in 1991).
For the data used in the estimations, the model predicted correctly between 50% and 88%
of the observed responses, depending on the sector–region combination. For most groups
the model predicted correctly about 70% of the responses. With the estimated parameters
we calculated the predicted probabilities of entering in 1992 (exiting in 1991) using the
data for plants that in our database appeared as entries of 92 (exits of 91). Out of these
groups, we kept in our database only the 179 (372) plants that presented highest predicted
probabilities of entering (exiting). The number of plants we actually kept in the sample
corresponds to the average of entries in 1991 and 1993 (exits in 1990 and 1992).
References
Aghion, Philippe, Howitt, Peter, 1992. A model of growth through creative destruction. Econometrica 60 (2),
323–351.
Aw, Bee-Yan, Chen, Siaomin, Roberts, Mark, 1997. Firm-level evidence on productivity differentials, turnover
and exports in taiwanese manufacturing. NBER Working Paper 6235.
Bartelsman, Eric, Dhrymes, Phebus, 1998. Productivity dynamics: U.S. manufacturing, plants 1972–1986.
Journal of Productivity Analysis 9 (1), 5–34.
Bartelsman, Eric, Doms, Mark, 2000. Understanding productivity: lessons from longitudinal microdata. Journal
of Economic Literature 38 (2), 569–594.
Bartelsman, Erik, Caballero, Ricardo, Lyons, Richard, 1994. Customer and supplier-driven externalities.
American Economic Review 84 (4), 1075–1084.
Bartelsman, Eric, Baily, Martin, Haltiwanger, John, 2001. Labor productivity: structural change and cyclical
dynamics. Review of Economics and Statistics 83 (3), 420–433.
Baily, Martin, Hulten, Charles, Campbell, David, 1992. Productivity dynamics in manufacturing establishments.
Brookings Papers on Economic Activity: Microeconomics, 187–249.
Basmann, Robert, 1960. On finite sample distributions of generalized classical linear identifiability test statistics.
Journal of the American Statistical Association 55, 650–659.
Basu, Susanto, Fernald, John, 1997. Returns to scale in U.S. production: estimates and implications. Journal of
Political Economy 105 (2), 249–283.
Bernard, Andrew, Jensen, Bradford, 1999. Exceptional exporter performance: cause, effect, or both? Journal of
International Economics 47 (1), 1–38.
Bernard, Andrew, Eaton, Jonathan, Jensen, Bradford, Kortum, Samuel, 2003. Plants and productivity in
international trade. American Economic Review 93 (4), 1268–1291.
Bigsten, Arne, Collier, Paul, Dercon, Stefan, Gauthier, Bernard, Gunning, Jan, Isaksson, Anders, Oduro, Adena,
Oostendrop, Remco, Pattillo, Cathy, Soderbom, Mans, Sylvain, Michel, Teal, Francis, Seufack, Albert, 1997.
Exports and Firm-Level Efficiency in the African Manufacturing Sector. University of Montreal, Montreal.
Burki, Shajid, Perry, Guillermo, 1997. The Long March: A Reform Agenda for Latin America and the Caribbean
for the Next Decade. World Bank, Washington, DC.
Caballero, Ricardo, Hammour, Mohamad, 1994. The cleansing effects of recessions. American Economic Review
84 (5), 1356–1368.
Caballero, Ricardo, Hammour, Mohamad, 1996. On the timing and efficiency of creative destruction. Quarterly
Journal of Economics 111 (3), 805–852.
Chen, Tain-jy, Tang, De-piao, 1987. Comparing technical efficiency between import-substitution-oriented and
export-oriented foreign firms in a developing country. Journal of Development Economics 36, 277–289.
Clerides, Sofronis, Lach, Saul, Tybout, James, 1998. Is learning by exporting important? Micro-dynamic evidence
from Colombia, Mexico and Morocco. Quarterly Journal of Economics 113 (3), 903–947.
Davis, Steven, Haltiwanger, John, Schuh, Scott, 1996. Job Creation and Destruction. MIT Press, Cambridge,
Mass.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371370
Edwards, Sebastian, 2001. The Economic and Political Transition to an Open Market Economy: Colombia.
OECD, Paris.
Ericson, Richard, Pakes, Ariel, 1995. Markov-perfect industry dynamics: a framework for empirical work.
Review of Economic Studies 62 (1), 53–82.
Foster, Lucia, Haltiwanger, John, Krizan, Cornell, 2001. Aggregate productivity growth: lessons from
microeconomic evidence. In: Dean, Edward, Harper, Michael, Hulten, Charles (Eds.), New Developments
in Productivity Analysis. Univ. Chicago Press, Chicago.
Foster, Lucia, Haltiwanger, John, Syverson, Chad, 2003. Reallocation, Firm Turnover, and Efficiency: Selection
on Productivity or Profitability? University of Maryland, Mimeo.
Hall, Robert, 1990. Invariance properties of Solow’s productivity residual. In: Diamond, Peter (Ed.), Growth/
Productivity/Unemployment. MIT Press, Cambridge.
Haltiwanger, John, 2000. Aggregate Growth: What Have We Learned from Microeconomic Evidence? University
of Maryland, Mimeo.
Harrison, Ann, 1994. Productivity, imperfect competition, and trade reform: theory and evidence. Journal of
International Economics 36, 53–73.
Hopenhayn, Hugo, 1992. Entry, exit, and firm dynamics in long run equilibrium. Econometrica 60 (5), 1127–1150.
Jovanovic, Boyan, 1982. Selection and the evolution of industry. Econometrica 50 (3), 649–670.
Katayama, Hajime, Lu, Shihua and Tybout, James R., "Why Plant-Level Productivity Studies are Often
Misleading, and an Alternative Approach to Inference" (April 2003). NBER Working Paper No.
W9617.
Klette, Tor Jakob, Griliches, Zvi, 1996. bThe inconsistency of common scale estimators when output prices are
unobserved and endogenousQ, Journal of Applied Econometrics, 11 (4), 343–361.
Kugler, Adriana, 1999. The impact of firing costs on turnover and unemployment: evidence from the Colombian
labor market reform. International Tax and Public Finance Journal 6 (3), 389–410.
Kugler, Maurice, 2000. The Diffusion of Externalities from Foreign Direct Investment: The Sectoral Pattern of
Spillovers. Essays on International Technology Diffusion, Chap. 1, PhD dissertation, UC Berkeley.
Kugler, Adriana, 2004. Wage-shifting effects of severance payments savings accounts in Colombia. Journal of
Public Economics (in press).
Kugler, Adriana, Kugler, Maurice, 2003. The labor market effects of payroll taxes in a middle-income country:
evidence from Colombia. CEPR Working Paper 4046.
Kugler, Maurice, Rosenthal, Howard, 2005. Checks and balances: an assessment of the institutional separation
of powers in Colombia. In: Alesina, Alberto (Ed.), Institutional Reforms in Colombia. MIT Press
(forthcoming).
Levinshohn, James, Petrin, Amil, 1997. Productivity, Industry Dynamics, and International Competition: a
Structural Econometric Approach. University of Michigan, Mimeo.
Liu, Lili, 1996. Entry-exit, learning and productivity change: evidence from Chile. Journal of Development
Economics 42, 217–242.
Lora, Eduardo, 1997. A decade of structural reforms in Latin America: what has been reformed and how to
measure it. Inter-American Development Bank Working Paper 348.
Lora, Eduardo, 2000. What makes reforms likely? Timing and sequencing of structural reforms in Latin America.
Inter-American Development Bank Working Paper 424.
Marschak, Jacob, Andrews, William, 1944. Random simultaneous equations and the theory of production.
Econometrica 12 (3), 143–172.
Melitz, Marc, 2003a. The impact of trade on intra-industry reallocations and aggregate industry productivity.
Econometrica 71 (5), 1695–1725.
Melitz, Marc, 2003b. Estimating Firm-Level Productivity in Differentiated Product Industries. Harvard
University, Mimeo.
Olley, Steve, Pakes, Ariel, 1996. bThe dynamics of productivity in the telecommunications equipment industry,QEconometrica, 64 (6), 1263–98.
Pack, Howard, 1984. Productivity, Technology, and Industrial Development: A Case Study in Textiles. Oxford
Univ. Press, New York.
Pombo, Carlos, 1999. Productividad Industrial en Colombia: Una Aplicacion de Numeros Indices. Revista de
Economia del Rosario.
M. Eslava et al. / Journal of Development Economics 75 (2004) 333–371 371
Roberts, Mark, Skoufias, Emmanuel, 1997. The long-run demand for skilled and unskilled labor in Colombian
manufacturing plants. Review of Economics and Statistics 79 (2), 330–337.
Roberts, Mark, Supina, Dylan, 1996. Output price, markups, and producer size. European Economic Review 40
(3–5), 909–921.
Roberts, Mark, Tybout, James, 1996. Industrial Evolution in Developing Countries: Micro Patterns of Turnover,
Productivity and Market Structure. Oxford Univ. Press, New York.
Roberts, Mark, Tybout, James, 1997. The decision to export in Colombia: an empirical model of entry with sunk
costs. American Economic Review 87 (4), 545–565.
Shea, John, 1993. Do supply curves slope up? Quarterly Journal of Economics 108 (1), 1–32.
Shea, John, 1997. The input–output approach to instrument selection. Journal of Business and Economic
Statistics 11 (2), 145–155.
Syverson, Chad, 2002. Price Dispersion: The Role of Product Substitutability and Productivity. University of
Chicago, Mimeo.
Syverson, Chad, 2003. Market Structure and Productivity: a Concrete Example. University of Chicago, Mimeo.
Tybout, John, 2000. Manufacturing firms in developing countries: how well do they do, and why? Journal of
Economic Literature 38 (1), 11–34.
Tybout, James, Westbrook, Daniel, 1995. Trade liberalization and dimensions of efficiency change in Mexican
manufacturing industries. Journal of International Economics 39 (1–2), 53–78.
top related